|
Computer Sciences Seminar Engineering Approach to Human-Machine Interactions
Esther Levin Abstract For several decades spoken language research has been aiming at building machines that would approach human-like skills. Indeed during the past few years we observed not only a tremendous improvement in the performance of prototypical conversational systems built and maintained by several research institutions, but also the development and deployment of robust commercial applications targeted at substituting human agents and touch-tone systems in high-volume telephone transactions. Although it is widely recognized that the success of speech recognition technology is due to the use of statistical modeling techniques in association with large corpora of samples, traditionally the statistical approach was not extended to the high-level, symbolic components of human-machine interaction systems. Rather, these components were generally treated as auxiliary, a mere post processing based on heuristics and handcrafted rules where consistency was often hard to maintain. In this talk I'll describe the application of the engineering, statistical approach (as opposed to the classical linguistic approach) to human-machine communication, and support its validity using statistical learning theory results. First I will describe the now widespread, traditional notion of Hidden Markov Models for speech recognition, then I will introduce less traditional example of statistical modeling techniques for symbolic processing, namely spoken language understanding and dialog strategy learning. I will conclude with a view at the future of research in the field of human machine interactions. Bio |